The first test on the new machine went quite well. The system is now storing 4000 percepts, and seems to be performing quite well. Following is a dump of all percepts (stacked on top of each other) after ~90,000 frames:
Here is the debug log from the test:
Spikes in rendering are due to over activation of percepts due to predictor output. Early on many percepts (thousands) can be activated before the network is sufficiently trained. The fix was to limit the number of percepts rendered to 300, if more than that many are activated, they are not rendered.